Case Study: a US-based conversational AI company achieves faster, lower-cost customer interaction processing with Toptal

US-based Conversational AI Company reduces inference costs and speeds customer interaction processing with Toptal

A US-based conversational AI company partnered with Toptal to overcome a significant challenge in processing high volumes of customer interactions. The company needed to accurately classify customer intent without the high latency and cost of using a large language model (LLM) for every single interaction. Furthermore, its business teams were dependent on technical staff for data insights, which slowed down their ability to answer critical questions about churn, retention, and agent performance.

Toptal architected a hybrid solution using an Airflow pipeline that employed BERT and unsupervised clustering for routine intent classification, only sending ambiguous interactions to an LLM. This maintained precision while drastically cutting inference costs. Toptal also built a self-service database agent that translated natural language questions into SQL, allowing non-technical teams direct access to data. As a result, the company achieved faster, lower-cost processing and empowered its business teams to gain immediate insights without relying on engineers.